Governed mental health AI
QiXMind lets mental health teams review AI drafts, track escalation and follow-up, and keep signed evidence of what was recorded. It is built on QiXHealth and designed for clinicians and services. It has no patient-facing interface.
Available for service evaluation. Real-patient use requires approved clinical procedures and reviewed notices for the applicable jurisdiction.
The four questions
The receipt records a fingerprint of the de-identified message sent to the model and the declared de-identification mode. Someone holding that exact message can check that it matches. The fingerprint does not show that de-identification was complete.
Every gate on the profile, in order, with whether it fired and at what severity. The matched words are never stored.
The role the signer selected (clinician or supervisor, never a name), when, and which fired gates were explicitly acknowledged. A fired gate that has not been acknowledged blocks sign-off.
When a gate escalates, a follow-up is due, within 24 hours by default. The follow-up deadline tracks a recorded obligation. Urgent concerns require action under your service's immediate-response protocol. The receipt records whether the follow-up was marked complete, by which role, on time or late, and whether it was marked overdue.
QiXMind records configured checks, clinician acknowledgements and follow-up events as signed evidence that anyone can check for alteration. This supports review and accountability over time. A valid signature shows the record is unchanged since our service signed it. It does not show that the checks caught every risk, that a clinical judgement was correct, or that care improved an outcome.
Safety gates
The checks look for specific wording in the model's output. They read only the output, not the text entered by the clinician. They do not assess risk, and they can miss risk that is worded differently or that the model leaves out, so they support clinical review and never replace it. A fired gate must be acknowledged before sign-off. If the model declines to write a draft or returns an empty one, the output is held and cannot be signed off.
An escalation creates a follow-up obligation with a deadline (24 hours by default; your clinical lead should confirm or change it before any real-patient use). Overdue follow-ups are marked on the Follow-ups page; QiXMind does not yet send notifications. A flag calls for human review; it does not set a risk level or decide whether a legal reporting duty applies. If the check cannot run, the output is held.
| Gate | Youth mental health | Adult mental health |
|---|---|---|
| Risk to self | Enabled | Enabled |
| Risk to others | Enabled | Enabled |
| Child protection | Enabled | Not enabled in this default profile |
| Imminent danger or reporting consideration | Enabled | Enabled |
The workspace
Draft a structured mental health safety review from notes or a referral: risk signals, protective factors, escalation and handover, follow-up and a review checklist.
A concise summary of a referral or first contact for the care team.
A governed clinical note for the care plan, reviewed before use.
Referrals to mental health services, crisis teams, psychology, alcohol and other drug services, and community support.
Drafts have not been clinically evaluated. Review every draft against the source.
The dashboard shows receipts, safety gates, follow-ups and recent activity from your own records. Evidence and follow-up pages let you download receipts, check their signatures and confirm follow-ups by role.
Jurisdictions
In Australia, mandatory reporting and mental health law differ by state and territory, so the acknowledgement shows the summary for your service's configured state or territory. New Zealand notes cover the Medical Council's AI guidance and the transition to the Mental Health Act 2026, which commences on 1 July 2028. All notices are drafts pending legal review.
See the jurisdiction notesConsent and disclosure
Consent templates for adult and youth mental health in New Zealand and Australia. Sign-off can record what the clinician states about whether AI use was disclosed to the patient and how, and can link a consent record. The receipt carries this as recorded, not as independently checked.
How sign-off worksClinician lane in QiXChat
Clinicians in entitled organisations can use the "Clinician consult: mental health" lane in QiXChat. Replies are designed to pass the same safety gates as the workspace and link to sign-off in QiXMind.
About the clinician laneIndependent verification
Receipts are signed with a private Ed25519 key and can be verified offline using the corresponding published public key. When keys are rotated, retired keys will stay listed so older receipts keep verifying. A signed receipt establishes integrity and provenance within its verification scope. It does not prove clinical correctness, complete capture, that every risk was flagged, or legal compliance.
How receipts are verifiedCommon questions
No. QiXMind is designed for clinicians and services. It has no patient-facing interface.
No. It drafts material for a clinician to review. It is not designed to diagnose or grade risk, it does not assign a risk level, and it makes no claim about clinical outcomes. Compulsory assessment and treatment decisions stay with authorised clinicians under the relevant Mental Health Act.
QiXMind is the mental health workspace built on QiXHealth. It uses the same governed engine, consent records, evidence and receipts, with its own dashboard, safety review workspace, evidence and follow-up pages.
New Zealand, and Australia by state and territory. Australian and New Zealand notices are drafts pending legal review and are for evaluation and synthetic demonstrations; reviewed notices are required before use with real patients. Each receipt records which notices were shown, by identifier.
Yes. Receipts are signed with a private Ed25519 key and can be verified offline using the corresponding published public key and the open source IOA Core tool, without contacting OrchIntel. Verification shows the receipt has not been altered since it was signed; it does not check the facts the receipt records.
The output is held and not shown, and the hold is recorded. A held output cannot be signed off or released. A regenerated draft is a new output and is checked again.
QiXMind runs in AWS us-east-1 (United States). Pseudonymised text is sent to model providers (OpenAI or Google); their processing locations have not yet been confirmed. Australian or New Zealand hosting is not offered today.
Not yet. It has been tested with synthetic data only. It has had no clinical validation, no evaluation across patient groups and no use with real patients.
Partners
We are exploring partner integrations for independently verifiable governance records around existing workflows. No integration is available yet.
Talk to us about partneringChoose adult or youth mental health, your country and, in Australia, your state. The intake sets up the default safety gates, follow-up rule and consent templates for your care setting, for your clinical lead to review before any real-patient use.
